Models of SARS-CoV-2 genomic elements generated by FARFAR2
We've organized the simulations we ran into directories according to the genomic region they came from: the 5′ or 3′ UTR or the frame shift element. Each simulation's subdirectory contains a pymol session (the exact session used for figure rendering) and a number of PDB files. These PDB files include all the cluster centers from the FARFAR2 clustering protocol, as well as up to 50 additional members of each cluster center. The cluster centers are numbered c.1.1.pdb through c.10.1.pdb, while the best-energy cluster's associated models are numbered c.1.1.pdb to c.1.50.pdb (if all 50 models are present).
As discussed in our manuscript, we ran several very large scale simulations (generating 500,000-4,000,000 models!) on the Open Science Grid, allowing us to sample some new structures very extensively. We organized these runs into the above system, but we prefixed all of the folders connected to Open Science grid runs with osg_. Similarly, we have revised some of our secondary structure predictions. For example, we now are interested in a combination of different secondary structures for the frameshift element (FSE), and we believe a shorter 5′ UTR SL4 and a longer 5′ UTR SL8 are more accurate models. Accordingly, we have renamed some existing folders to have the suffix _old as these models, while still valid, reflect secondary structure assumptions that are not current.
We have many collections of OSG models that compare different sets of secondary structure assumptions to each other. These simulations have infixes indicating the affiliated lab whose data inspired the predictions, discussed more in the manuscript.
Finally, we include pocket predictions from fpocket for the model sets here in the pocket_predictions folder. The pockets are represented in each PDB file by clusters of atoms with residue name STP and residue number identifying the pocket to which they belong.
However you want! Of course, our most urgent hope is that these models will be useful to drug development against SARS-CoV-2. But if you are training a new RNA scoring function (...which itself may someday be useful against new viral pandemics), please make use of these models. To benchmark computational methods for RNA-targeted drug discovery, you may wish to look at the models in FARFAR2-Apo-Riboswitch.